Vibe Marketing Tutorial: Automated Lead System With Claude AI + N8N + Airtable

Key Takeaways

Hey there. I'm going to show you an extremely powerful and very simple to set up workflow that allows you to conduct research on small and local business leads, extract a ton of valuable information, and have a pitch email automatically generated that you can send out to start getting clients right away. If you like this video, please like and subscribe. And you can visit the viableedge.com or sign up for the viable edge newsletter in the link below. I'm going to walk through the entire process of setting this up. There's a leads guide linked below that contains all of the files and documents that you'll need for this. That file will contain six folders. Each folder has the files that go with each step in this process to make it as easy to follow and set up as possible. I'm going to get into everything you need to set this up so you can get going right away. So, let me show you exactly how this works. I have a Claude project set up here and all I need to do is provide a business type and a location and Claude will do the rest. So, just thinking randomly, let's think about HVAC in Salt Lake City. Now, what happens here is Claude is going to go out and do research because Claude has built-in web searching capabilities. Once Claude finds leads that it qualifies according to criteria that I set in the prompt in the project folder, it will then add them to an air table. And once specific leads are found, if they meet a certain quality threshold and lead score, then a pitch email will automatically be generated and left in my Gmail drafts folder. Now, Claude has reported that it's found several possible candidates for leads. It's adding them to my air table. You can see on the bottom here, they're starting to show up in real time. We got technical HVAC design. And as these leads are dropped in, they're scored a certain way. So, we've got a seven and an eight. Because those meet or exceed a threshold that I set, pretty soon, we're going to see an email get written. That'll be a personalized pitch based on the issues we've identified with this business. If I click through to any one of these records, we can see that there's three main pieces of information that are gathered. A core issue. The website shows evidence of manual processes, etc. A modernization opportunity, implement a digital intake workflow, and an AI solution path. A lightweight AI powered intake system could pre-qualify leads. This information is based on the services that I offer, which are what the prompt is designed to look for. And each one of these leads has that information in it. So one of the side benefits of this process is that over time you have a whole bunch of data and gathered research on what these types of businesses might struggle with which will enable you to refine your offer and give you some really valuable research for marketing to these businesses. All right. So we can see that while I was speaking these emails were written and popped up. So they should be appearing in my Gmail folder. Let's see. All right, here's my Gmail. And you can see I've got the three drafts right there. If I open one of these, hi there. I came across technical HVAC design while researching businesses in your category based on what I saw. I won't read the whole thing, but you get the idea. This is a really, really powerful and fantastic process. You're going to be amazed how easy it is to set up. I'm going to walk you step by step through the entire thing. So again, I encourage you to like and subscribe, and I'd love to get your feedback in the comments. Likewise, if you run into any issues, want any help or ideas on how to improve this, please leave a comment, or reach out to me at adamviableedge.com. Let's get right into it. Let's take a highle look at what's happening here. First, we're going to use Claude to conduct research. Effectively, Claude is really doing all of the heavy lifting here. And then once it finds leads, it's using the MCP server just to send that data out to Air Table. And I think this also speaks to how powerful the MCP trigger is because as much offloading as we can do to Claude desktop simplifies everything that happens on the automation side of things. So once Claude does that research, identifies leads, it's going to make a call via the MCP trigger to N8N and through that call it's going to pass data along which N8N will then transfer to an additional workflow and that workflow will record the data as record in air table. Then a third workflow is going to pull air tableable every minute. So that means it's going to literally scan air tableable every minute for new records. And if it sees a record with a lead score of seven or higher, it's going to trigger an AI node to write an email and leave that email in my draft folder and also record the subject line and the message back to the record in Air Table. So that's it. There's four big steps here. And as I mentioned, the first step which is happening on cloud desktop is really doing all the heavy lifting. And one of the reasons that I want to emphasize this is because this is where the creativity is unlocked. Anything that you could think to prompt Claude to do where you might want to pull data off platform or run it through an N8 workflow, you can mimic this process essentially. So if there's one takeaway from this video, let it be that. Again, I'd like to remind you that there's a guide linked below that has all of the templates, prompts, and documentation that I'm going to walk through with each step. So, I encourage you to download that first and then use that to follow along if you want to build with me in real time. Again, if you have any questions, please drop those in the comments and I'll do my best to answer them as soon as possible. So, let's get right into it for real this time. All right, let's go over some assumptions and requirements for this workflow. You should have a baseline understanding of N8N. How to open and manipulate and edit nodes is probably enough. You should be comfortable editing text files. This is used to edit the MCP server to communicate between Claude and NADN. If that sounds confusing, don't worry. I promise it'll be cleared up and it's easier than it sounds. Next, you should have a basic familiarity with spreadsheets. I'm using Air Table for this, but you can use Google Sheets if you want. Finally, you should have some level of understanding of how to get and use an API key for AI. The AI API is used two times in the workflow structure. There's one node to sort and structure data, and there's another node to write a pitch email. Other than that, Cloud Desktop will be doing the heavy AI lifting. So, aside from those basic assumptions, this process will use Claude Pro. So, you'll need a paid Claude account. It does work with free, but you'll run through your credits very fast. So, I definitely recommend Pro. You'll want an N8 account, which is also paid, but they have a free trial. There are ways to self-host N8. I'm not going to get into that in the video, but if you want to research how to use N8N for free, you can do that, and there's plenty of resources out there. You'll need a Google Docs account which is free as well as Google Drive. Google Drive is what integrates with Claude and provides that visibility to Google Docs. So that's all within one Google account. You'll want an Air Table account. The free account is enough. Gmail, what is used at the tail end of the workflow to capture the draft pitch emails. And as I mentioned earlier, you're going to want an AI API key. Anthropic, OpenAI, Google. You can also go to Open Router. There's a lot of options out there and it's pretty straightforward. Linked below is a file that has all of the templates and docs. It'll have a written guide and then it'll have folders for each step and each folder will contain the files that are used in that step. So, it should be as easy to follow as possible. Keep your eye on the file names. I may have changed some things around as I go through the work of editing this. And it's important to know the file names of the Google Docs in respect to the project instructions. And I'll get into that a little bit later. And that's pretty much all you'll need to get going. Let's move on to step one. In the lead research materials that's linked below, you'll find that every file you need is organized by step. So, if we open up step one, there's going to be five different document files, and you should import all of these into Google Docs. I have them all imported on the left side of the screen here and I'll walk through each file. First, we have our claude instructions. This is what the project is going to use as a general guide. We have our prompt. So, the prompt will live as a doc in the project. That way, you don't have to copy and paste it every time you want to use it and you can update it in Google Docs and Claude will get those updates dynamically. Next is the email prompt. This email prompt is not actually used by the project, but it is here as a reference and for you to be able to edit and refine in Google Docs. This is used later within NADN. Next, we have an AI modernization prospects guide. And all this really does is lay out the criteria for the best fitting businesses for the type of service that I want to offer. So, I encourage you to look at this framework and edit it, adjust it, and refine it for how it's going to work best for your business and use AI for this. But it's a great way to keep Claude focused on the types of businesses that you want to remain focused on. And finally, there's a priority matrix which has the specific details based on the last document. So, this goes into specific cities and businesses and lays out by priority which businesses are most likely to respond to my offer. Once you have these docs in Google Docs, start a new project in Claude. You can name your project whatever you want. Let's just call it SMB research for now. And then once this project is created, you want to set your project instructions. Copy and paste the entire instructions from the Word doc into Claude's project. and then save the instructions. We're going to come back and revisit this later. And one thing I want to highlight right now for your awareness is that there is a key material section. And this key materials section references all of the other files so Claude knows what to use and when. So, it's important that if you change your file names that those updates are reflected in the key materials section. The next step is adding your Google Docs into the cloud project. If you haven't already, make sure that you connect your Google Drive account with Claude. There's a simple integration procedure in the settings. And once that's connected, if you go to the project knowledge and add files, you should see the docs show up in the list. And if you can't find them, you can search for them using this bar. Simply add each file in and they will take a minute to load in. And you'll see that your project context starts to fill up. Oh, I accidentally added the instructions. You don't have to add the instructions as a file. Now your project is set up and that completes step one. All right, let's move on to step two. This is setting up our air table. So if we open up the step two folder in the included file, you'll see that there's a CSV template there that has all the data schema or simply the columns and data fields that we'll need to capture records from cloud. So to get started, we're simply going to create or use a workspace and then start a new sheet from scratch. Let's import our CSV file. Give that a minute to load up. Once that's loaded up, there's a couple of things we want to do here. One is we want to go to the created time field, edit that field, and change the type to created time, which is down towards the bottom of the list. What this does is tell Air Table to capture the date and time anytime a new record is created. When NAND goes to pull air table later on for pitch emails, it can see what has been created since its last check and run the evaluation on whether that record gets email or not. The CSV file has some records in there. So if this happens, you see a bunch show up, you can delete them. Next, change the score field to a number so it matches the N8 workflow and when data is sent back and forth, it's properly characterizing the data types. The other fields should be long form text fields and they're set that way when the file is imported. Just want to note one thing. There's a field here for generate message. You can ignore that field for now. I use it to manually trigger emails. I'm not going to include that process in the video, but if there's enough interest, let me know in the comments. I'm happy to make a follow-up video to show you how to make the manual trigger. One thing to note is that you should rename your base and table to something that is going to make sense to you because we're going to reference these items in the N8N workflow. And that's it for step two. Moving on to step three, we're going to start working in N8N and set up our workflows. The first workflow to set up is the capture lead to air table workflow, which is actually second in the process after the MCP trigger. We want to set this up first so that when we do set up the MCP trigger, we can point it to this workflow. The file for this workflow can be found in the step three folder of the guide linked below. To import this file, simply create a new workflow within N8N. Click on the three dots on the top right of the screen, import file, and then choose the file that's in that folder. Once you have that file imported, you should see something like what you see on the screen here. So, taking a look at these nodes. First, we're going to look at Air Table. And if you haven't already, you're going to want to connect your Air Table account to N8N. And you can do that by creating a new credential. In Air Table, you get a personal access token, which you use just like an API key. If you can't figure it out, Google or ask chatgbt. It's actually very easy to do. Once you have your Air Table account connected, you'll see that these values base and table, you want to pick from the list base and the table that you created in the prior step. These values should already be present in the included import. If they're not for some reason, drop a comment and I'll see what I can do to help you out. Next, let's look at the middle node, which is used to sort and structure data. the data is going to be coming in via the MCP trigger as one long string which is why you have this blob of text here as a query. So what this node does is take that information and then break that out into the values that can be picked up by air table. So if I run this test here, you can see that in action. I've executed this note already, so there's some data already there for me to test with. So it might be a little bit hard to see. I apologize. So now all of that data is structured and organized by the schema we set and matches the column headers in air table. Business name, URL, email score, which is a number, core issue, modernization, opportunity, and AI solution pad. So all of this will now be transferred over to the air table node and dropped into the appropriate columns when a new record is created. And the way that it's doing this is through AI. So this is the first place you're going to use your AI API key. If we click down here, I already have Anthropic hooked up. I'm using Claude Sonnet 3.5, which is probably even a little bit overkill for this process, but once you import this workflow and connect your AI model, you should be set to have the raw data structured and organized accordingly. One thing I also want to note here is that this workflow will actually remain inactive because it's being called and activated by another workflow. It doesn't need to be active on its own. And that's it. That's all you need to do to set this up. So, make sure you save. And then for step four, we'll start another new workflow. Moving on to step four, the MCP server trigger. This is what allows Claude to communicate via MCP to N8N and send data and also receive data. It's a powerful capability that's surprisingly easy to set up. Similar to the last step, if you go into the step four folder of the guide linked below, click on the three dots to import that JSON file. And once you import that into the new workflow, you should see what we're looking at on the screen here. Once you have this loaded, open up the MCP server trigger and capture the production URL. You can flip back and forth between the test and production URL. It's important to capture the production URL. Also notice the path. The path is set as SMB research adds that matches the path at the end of the URL. Also note that the production URL does have an additional path after the slash SSE. So, I want you to capture the production URL, copy it into a notepad. We're going to use this later when we set up our MCP server. The second node is our tool call. This one is called add lead to air table. And this is what sends the data to the workflow that we just set up. And the reason that we set up that workflow first was so that when we set this one up, we have that workflow already visible in the workflow list. So, make sure that the source is set to database, that the workflow parameter is from list, and then all you need to do is select the workflow that we just set up. One thing to note is that the name of this tool is what Claude will see as the MPC server when you're using it. So, if you noticed when I showed the demo up front in the chat box, when I referenced the tool, it matched what I have written here. So, I'm hopeful that makes sense. That is the entire setup for step four. So, make sure you save and we'll move on to the next step. Moving along, step five is where we set up our workflow for drafting a pitch email. Similar to the prior steps, if you open up the step five folder in the linked lead guide, you'll find the N8N template that you could import directly into a new workflow. This one's a little bit more complex than the prior ones we looked at, but easy to set up. The way that this works is that there's an air table trigger that is set to scan the air table every minute and any time that it finds a new record, it will make a determination if that record is eligible for a draft email. In this trigger, we're pointing to our air table via the base and table ids. These values are a string of characters that can be found in the air table URL. For one reason or another, you can't designate the air table information by name. So the base ID is the string of characters that appears after the first slash in the URL and starts with the letters a and then the table ID is the string of characters that are after the following slash. So all you need to do is copy and paste those two values into the base and table ID fields here in the trigger. Additionally, if you want, you can change how frequently N8N will pull or scan Air Table. By default in this workflow, I have it set for 1 minute. This is also why it was important to set up that created time column because that's where this trigger is looking. Anytime it sees a new value in the created time column that it has not previously analyzed, it'll look at that record, determine if it needs an email, and then execute the workflow. All right. So once air table is triggered the next node is this if node and here we determine that if there's a score that is greater than six so 7 8 9 or 10 will signal true and then pass the data on to the email generation and if it's a six or below it'll signal false and then end the operation. So if that score is higher than a six in a new record the email generation node is triggered. And this is probably the most complex node that we have in the entire process. In this field here, I just expanded the prompt field. This prompt is found in the email prompt Google Doc that we covered in step one. So, all you really need to do is copy and paste that email prompt in. But before you do, you have to customize the email because this is just a default setup that I have here. I have included my name. You'll see that there's different fields referenced in here. I encourage you to take a look and craft the email prompt that is going to be a great fit for you. Additionally, as I have used this and tested this process, this is where I find myself doing the most fine-tuning and refinement. So, normally I'll do the work in the Google doc and copy and paste the prompt into here when I'm ready to test it again. If that's confusing or if you need help with this step, please leave a comment and I'll do what I can to help out. Another thing to note is that the prompt instructs the AI to write the email in HTML. And the reason that I did this was because I like to include my signature in a certain way. And HTML allowed me to do that. So you can have the email written in text or HTML. And if you don't know how to write HTML, and I don't, just use a chat app to help you out. Similar to the prior step, we are hooked up to the anthropic model for our LLM. And I'm using Claude 3.7 Sonnet here because I like its writing capabilities, but you can use any model you like. Additionally, there's a tool that structures the output into an email format for Gmail and then for Air Table to record the subject line and message back into the appropriate record just to have it as a reference. So similar to the prior step, you can set up the air table node at the end of the process to point to the air table. In this one, you can choose the name of the base and the table from the list. And then the other items should be set up via the import. For Gmail, you want to make sure that your Gmail account is connected through credentials. This uses Google's OOTH, so all you need to do is log into your Gmail. When you add a new credential, it'll prompt you to sign in with Gmail. And then you'll see that the subject and message values are dynamically calling what is the output from the prior email node. The resource is draft and the operation is create. And that's what tells this node to leave a draft in your Gmail draft folder. So, I'm hopeful that this is clear. The instructions included in the guide will go more into some of the specific details, but if you understand how this workflow works, it should be very easy to set up. Make sure that this workflow is set to active. And if you're testing or if you're testing workflows prior in the process and you don't want to activate this because you are testing the research prompt, for example, deactivate it. Otherwise, you might just trigger a whole bunch of emails you'll never end up using. And that's it for this step. Moving along to step six, we're going to set up the MCP server to allow Claude to communicate with N. MCP is simply a protocol that allows Claude to access and use a variety of databases and tools and platforms without the need for specialized API connections or endpoints. So the first step here is to have Claude open and go into your settings. There are two settings areas in Claude. One is the account settings which is on the bottom left and the other one is your developer app settings on the top left on Mac. So from the top menu here you can access them and on Windows I think it's on the hamburger menu but it's not hard to find. What we're looking for are the developer settings. So once you open this setting window you'll see there's a developer tab. Click on that developer tab and then we're going to click on edit config. Once you click on edit config, a folder will open on your computer. And if you don't have one already, there will be a file called claw desktop config.json. Open that file with any text editor. So once you open this up with a text editor, you may not see anything. I have some MCP servers already installed here, including the one we're looking at. This file might be blank, but what you need to do is open the folder for step six in the linked guide. Whoops. We're going to open this with a text editor, too. We're going to want to copy and paste the entirety of this code into the claw desktop config JSON. Now, if you have JSON code in here already, it means you probably have an MCP server installed, so you've gone through this process and you should know how to add this. But if you don't, all you need to do is copy and paste the entirety of this code over. But there's one really important thing we need to do here. If you remember back in the earlier step, I think it was step four where we captured our production URL from the MCP server trigger, we want to take that URL and we want to paste it in in the placeholder that I've highlighted here. This is what tells Claude to trigger that MCP's workflow, which then triggers the following workflow to record the record to Air Table. Now, once you've copied this MCP server code in, you want to save your desktop file. close out of it and then we're going to restart Claude. Once Claude is restarted in this search and tools button, you should now see N8 and leads along with add to air table once you click through appear. And this indicates that the MCP server has successfully installed. If you're running into errors or the MCP code isn't working, use chat GPT or Claude or any LLM to help you troubleshoot and problem solve the MCP server code. It's simple JSON and I've done that many times to get this working properly. Another way to check is to go back into the developer settings and you can see that there's a list of servers that are running and if I click on N8 and leads it indicates that it is running and then has some information about how to execute this server when it's called. And that's it for this step. Now that you know how to connect claw to N8N, there are unlimited things that you can do and it's an extremely powerful capability and this is where your creativity can really come in and you can really come up with some great applications and use cases for using Claude in new ways. So there will be more detailed instructions in the guide which you might have noticed I'm scrolling through on the left side over here. And if you have any trouble or you're running into issues, please drop a comment and I'll do whatever I can to help out. But hopefully you understand how easy it is to get Claude connected to N8N and begin leveraging a whole new dimension of capabilities right away. With this step complete, we've really done all of the hard work. The next thing to do is to start testing, refine our prompts, and fine-tune the process to work for you. With everything set up, it's a good opportunity to go back into the clawed project instructions, which I have pulled up here in the Google Doc, and make sure that the key materials are specifically referencing the files that are in the claude project. It's easier to work in Google Docs. That's just how I prefer to do it. And then copy and paste this into the claude instructions once the file names match. So, I'll show you what that looks like in my project. So, if you recall in the first step when we set up our project, we pasted the project instructions in here from the Google doc. And if we scroll down to key materials, we have 1, two, three, four files. The lead research prompt, the email prompt, the priority matrix, and the prospecting guide. Each name here should be a clear reference to each of the four files that are set up in the project. Make sure that you double check that everything matches up. And that's what allows me to enter just the business and the location because once the chat sees that and references the instructions, it knows then to execute the project research prompt. I mentioned a couple times throughout the video, you can go back and fine-tune the research prompt as you determine the quality of the results that you're getting. So, I highly encourage you to take some time, open up that prompt document, and spend some time going through the different criteria to make sure that this matches the kinds of services that you're selling, that the criteria match what you're looking to find in prospective leads and fit the nature of the research and the outcome that you're looking for. So, I wanted to just give you a quick tip on what to do if you're running into errors. The first thing you're want to do is go to executions. This is going to document all of the previous executions of the workflow and enable you to go and recall the data that was received and use that data to edit out any issues. So, if I go to the most recent workflow, which is what I used in the beginning of this video, and I click through to the node, let's click through to the tool node. Actually, the query that's received here is what Claude sent through. And what we're looking at on the right is what was returned from the other tool. One thing I want to call out here is that this ID is what is assigned by air table in the record once that record is created. So if we look at the lead capture workflow and go into the most recent execution, we can see that the string of data came in from the MCP trigger. We can see that that data was then sorted according to our instructions and then it was sent to air table as records. Now all of the values here they're lit up green meaning that this node was able to associate the data with the proper variable that was set. So for example in our schema the business name is I know a guy back. The website URL is assigned as I know a guy back. So this is where things usually break down. If you're not getting the proper data in Air Table or you're seeing errors thrown up, the first place to look is to make sure the variables are landing in the right spot. And what you can do if you need to edit this is you can copy the data to editor. So what this does is it uses the most recent execution as the sample data for you to edit the workflow. So, if I go into Air Table now, I can edit it with the same data and I can pull these variables into the proper fields if I need to. I'm not going to do that because they're already set up properly. Make sure schema is selected here. That'll turn these variables into dragable elements that you could literally drag into the proper field. So, you might have to poke at this for a little while until you get the variables working properly. So, again, if you need troubleshoot, start there. If you are troubleshooting with the research prompt, I would encourage you to deactivate Mail Prompt. That way, as you're testing and refining the research, you're not just generating emails, which could be wasting valuable API credits. And the last tip I want to share is make sure that whatever email address you're sending from is following best practices and guidelines for deliverability. You can go research this on Google or NS ChatgBT. There are settings that you might need to change to ensure that your emails are going to be deliverable and not set up to trigger spam filters, which could really damage your domain or your brand. Well, thank you so much. I really hope that you have as much fun as I did tinkering, experimenting, coming up with new and creative ways to leverage technology for Vibe Marketing. People like you and I can now do things that we couldn't do even 6 months ago. We have the power of entire teams now at our fingertips and we're only limited by our own creativity. If you like this video, again, please give me a like and subscribe. It really goes a long way. Sign up for the viableedge.com. And if you have any feedback or would like any assistance or there's parts of the process that are not clear enough, leave a comment. I'm happy to go back and figure anything out with you. Thank you so much again and thank you so much for sticking with this. Bye-bye. [Music] Yeah. Oh yeah.